Files
The-Art-of-Linear-Algebra/The-Art-of-Linear-Algebra.tex
T

719 lines
20 KiB
TeX
Raw Normal View History

2021-09-02 11:31:08 +09:00
\documentclass[letterpaper]{article}
\usepackage{typearea}
\typearea{12}
\usepackage{here}
\usepackage{bm}
\usepackage{amsmath, amsfonts}
\usepackage[top=20truemm,bottom=20truemm,left=25truemm,right=25truemm]{geometry}
\usepackage[dvipdfmx]{hyperref,graphicx}
2022-07-23 23:02:32 +09:00
% incorporated from Linear Algebra for Everyone 7/18/2022
\newcommand{\bi}[1]{\hbox{\boldmath$#1$}}
\DeclareRobustCommand\transp{^{\mathrm{T}}}
\DeclareMathAlphabet{\cmrv}{OML}{cmm}{b}{it}
\newcommand{\bu}{\hbox{\boldmath$u$}}
\newcommand{\bv}{\hbox{$\cmrv{v}$}}
\newcommand{\bw}{\hbox{\boldmath$w$}}
\newcommand\mat{{\sf MATLAB}}
%
2022-08-14 15:51:43 +09:00
% prepare to move figures
\graphicspath{ {figs/} }
2021-09-02 11:31:08 +09:00
\begin{document}
2025-06-30 14:22:41 +09:00
% Manual title block with numbered footnotes
\begin{center}
{\LARGE The Art of Linear Algebra\\
2021-09-02 11:31:08 +09:00
\vspace{5pt}
\large{
-- Graphic Notes on ``Linear Algebra for Everyone" --
}
}
2025-06-30 14:22:41 +09:00
\vspace{1.5em}
Kenji Hiranabe\footnote{twitter: @hiranabe, k-hiranabe@esm.co.jp, \url{https://anagileway.com}} \\
with the kindest help of Gilbert Strang\footnote{Massachusetts Institute of Technology, \url{http://www-math.mit.edu/\~gs/}}
2021-09-02 11:31:08 +09:00
2025-06-30 14:22:41 +09:00
\vspace{1em}
September 1, 2021/updated \today
\end{center}
2021-09-02 11:31:08 +09:00
2025-06-30 14:22:41 +09:00
\vspace{1.5em}
2021-09-02 11:31:08 +09:00
\vspace{-5pt}
2023-09-10 17:03:13 -04:00
2021-09-02 11:31:08 +09:00
\begin{abstract}
2023-09-10 17:03:13 -04:00
I try to intuitively visualize some important concepts introduced
in ``Linear Algebra for Everyone",\footnote{``Linear Algebra for Everyone":
2023-09-12 16:58:18 +09:00
\url{http://math.mit.edu/everyone/} with Japanese translation from Kindai Kagaku.}
2023-09-10 17:03:13 -04:00
which include Column-Row ($\bm{CR}$), Gaussian Elimination ($\bm{LU}$),
2022-08-14 15:51:43 +09:00
Gram-Schmidt Orthogonalization ($\bm{QR}$), Eigenvalues and Diagonalization ($\bm{Q \Lambda Q\transp}$),
2023-09-10 17:03:13 -04:00
and Singular Value Decomposition ($\bm{U \Sigma V\transp}$).
This paper aims at promoting the understanding of vector/matrix calculations
2023-09-12 16:58:18 +09:00
and algorithms from the perspective of matrix factorization.
2023-09-10 17:03:13 -04:00
All the artworks including the article itself are maintained under the GitHub repository \url{https://github.com/kenjihiranabe/The-Art-of-Linear-Algebra/}.
2021-09-02 11:31:08 +09:00
\end{abstract}
\section*{Foreword}
I am happy to see Kenji Hiranabe's pictures of matrix operations in linear algebra !
The pictures are an excellent way to show the algebra. We can think of matrix
multiplications by row $\bm{\cdot}$ column dot products, but that is not all -- it is ``linear combinations"
and ``rank 1 matrices" that complete the algebra and the art.
I am very grateful to see the books in Japanese translation
and the ideas in Kenji's pictures.
\begin{flushright}
-- Gilbert Strang \\ Professor of Mathematics at MIT
\end{flushright}
\tableofcontents
\section{Viewing a Matrix -- 4 Ways}
2023-09-10 17:03:13 -04:00
A matrix ($m \times n$) can be viewed as $1$ matrix, $mn$ numbers, $n$ columns and $m$ rows.
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
\centering
2022-10-15 15:56:30 +09:00
\includegraphics[scale=0.8]{ViewingMatrix-4Ways.eps}\\
2022-08-13 12:37:44 +09:00
\caption{Viewing a Matrix in 4 Ways}
2021-09-02 11:31:08 +09:00
\end{figure}
\begin{equation*}
A= \begin{bmatrix}
a_{11} & a_{12}\\
a_{21} & a_{22}\\
a_{31} & a_{32}
\end{bmatrix}
=
\begin{bmatrix}
| & |\\
\bm{a_1} & \bm{a_2}\\
| & |
\end{bmatrix}
=
\begin{bmatrix}
- \bm{a_1^*} -\\
- \bm{a_2^*} -\\
- \bm{a_3^*} -
\end{bmatrix}
\end{equation*} \\
Here, the column vectors are in bold as $\bm{a_1}$.
Row vectors include $\bm{*}$ as in $\bm{a_1^*}$.
Transposed vectors and matrices are indicated by $\mathrm{T}$ as
2022-07-23 23:02:32 +09:00
in $\bm{a}\transp$ and $A\transp$.
2021-09-02 11:31:08 +09:00
\section{Vector times Vector -- 2 Ways}
2023-09-10 17:03:13 -04:00
Hereafter I point to specific sections of ``Linear Algebra for Everyone"
2021-09-02 11:31:08 +09:00
and present graphics which illustrate the concepts with short names
2023-08-22 09:07:48 +09:00
in gray circles.
2021-09-02 11:31:08 +09:00
\begin{itemize}
\item Sec. 1.1 (p.2) Linear combination and dot products
\item Sec. 1.3 (p.25) Matrix of Rank One
\item Sec. 1.4 (p.29) Row way and column way
2022-10-15 14:45:45 +09:00
\end{itemize}
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
2022-08-13 12:37:44 +09:00
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.6]{VectorTimesVector.eps}
2021-09-02 11:31:08 +09:00
\caption{Vector times Vector - (v1), (v2)}
\end{figure}
2023-09-10 17:03:13 -04:00
(v1) is an elementary operation of two vectors, but (v2) multiplies the column to the row
and produces a rank 1 matrix. Knowing this outer product (v2) is the key to the following sections.
2022-08-14 15:51:43 +09:00
2021-09-02 11:31:08 +09:00
\section{Matrix times Vector -- 2 Ways}
A matrix times a vector creates a vector of three dot products (Mv1)
as well as a linear combination (Mv2) of the column vectors of $A$.
\begin{itemize}
\item Sec. 1.1 (p.3) Linear combinations
\item Sec. 1.3 (p.21) Matrices and Column Spaces
2023-09-10 17:03:13 -04:00
\end{itemize}
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.6]{MatrixTimesVector.eps}
2021-09-02 11:31:08 +09:00
\caption{Matrix times Vector - (Mv1), (Mv2)}
\end{figure}
At first, you learn (Mv1). But when you get used to viewing it as (Mv2),
2022-09-13 11:51:43 +09:00
you can understand $A\bm{x}$ as a linear combination of the columns of $A$.
2021-09-02 11:31:08 +09:00
Those products fill the column space of $A$ denoted as $\mathbf{C}(A)$.
2022-09-13 11:51:43 +09:00
The solution space of $A\bm{x}=\bm{0}$ is the nullspace of $A$ denoted as $\mathbf{N}(A)$.
2023-08-22 09:07:48 +09:00
To understand the nullspace, let the right-hand side of (Mv1) be $\bm{0}$
and see all the dot products are zero.
2021-09-02 11:31:08 +09:00
2023-09-10 17:03:13 -04:00
Also, (vM1) and (vM2) show the same pattern for a row vector times a matrix.
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.6]{VectorTimesMatrix.eps}
\caption{Vector times Matrix - (vM1), (vM2)}
\end{figure}
The products fill the row space of $A$ denoted as $\mathbf{C}(A\transp)$.
2023-08-22 09:07:48 +09:00
The solution space of $yA=0$ is the left-nullspace of $A$, denoted as $\mathbf{N}(A\transp)$.
2023-09-10 17:03:13 -04:00
The four subspaces consist of $\mathbf{N}(A)$ + $\mathbf{C}(A\transp)$
2021-09-02 11:31:08 +09:00
(which are perpendicular to each other) in $\mathbb{R}^n$ and
2022-08-14 15:51:43 +09:00
$\mathbf{N}(A\transp)$ + $\mathbf{C}(A)$ in $\mathbb{R}^m$
2021-09-02 11:31:08 +09:00
(which are perpendicular to each other).
2021-09-02 11:31:08 +09:00
\begin{itemize}
\item Sec. 3.5 (p.124) Dimensions of the Four Subspaces
2023-09-10 17:03:13 -04:00
\end{itemize}
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.8]{4-Subspaces.eps}
2021-09-02 11:31:08 +09:00
\caption{The Four Subspaces}
\end{figure}
See $A=CR$ (Sec 6.1) for the rank $r$.
\section{Matrix times Matrix -- 4 Ways}
``Matrix times Vector" naturally extends to ``Matrix times Matrix".
\begin{itemize}
\item Sec. 1.4 (p.35) Four Ways to Multiply $\bm{AB=C}$
\item Also see the back cover of the book
2023-09-10 17:03:13 -04:00
\end{itemize}
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.6]{MatrixTimesMatrix.eps}
2021-09-02 11:31:08 +09:00
\caption{Matrix times Matrix - (MM1), (MM2), (MM3), (MM4)}
\end{figure}
\section{Practical Patterns}
Here, I show some practical patterns which allow you to capture
2023-09-10 17:03:13 -04:00
the upcoming factorizations in a more intuitive way.
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.6]{Pattern12.eps}
2021-09-02 11:31:08 +09:00
\caption{Pattern 1, 2 - (P1), (P1)}
\end{figure}
Pattern 1 is a combination of (MM2) and (Mv2).
2023-09-10 17:03:13 -04:00
Pattern 2 is an extension of (MM3). Note that Pattern 1 is a column operation (multiplying a matrix from right),
2021-09-02 11:31:08 +09:00
whereas Pattern 2 is a row operation (multiplying a matrix from left).
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.6]{Pattern11-22.eps}
2021-09-02 11:31:08 +09:00
\caption{Pattern 1$^\prime$, 2$^\prime$ - (P1$^\prime$), (P2$^\prime$)}
\end{figure}
2023-09-10 17:03:13 -04:00
(P1$^\prime$) multiplies the diagonal numbers to the columns of the matrix,
whereas (P2$^\prime$) multiplies the diagonal numbers to the row of the matrix.
2021-09-02 11:31:08 +09:00
Both are variants of (P1) and (P2).
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.6]{Pattern3.eps}
2021-09-02 11:31:08 +09:00
\caption{Pattern 3 - (P3)}
\end{figure}
2023-09-10 17:03:13 -04:00
This pattern emerges when you solve differential equations and recurrence equations:
2021-09-02 11:31:08 +09:00
\begin{itemize}
\item Sec. 6 (p.201) Eigenvalues and Eigenvectors
\item Sec. 6.4 (p.243) Systems of Differential Equations
2023-09-10 17:03:13 -04:00
\end{itemize}
2021-09-02 11:31:08 +09:00
\begin{align*}
\frac{d \bm{u}(t) }{dt} &= A \bm{u}(t), \quad \bm{u}(0)=\bm{u}_0\\
\bm{u}_{n+1} &= A \bm{u}_n, \quad \bm{u_0} = \bm{u}_0
\end{align*}
In both cases, the solutions are expressed with
2023-09-10 17:03:13 -04:00
eigenvalues ($\lambda_1, \lambda_2, \lambda_3$),
2021-09-02 11:31:08 +09:00
eigenvectors $X=\begin{bmatrix} \bm{x}_1 & \bm{x}_2 & \bm{x}_3 \end{bmatrix}$ of $A$, and
2022-08-14 15:51:43 +09:00
the coefficients $c=\begin{bmatrix} c_1 & c_2 & c_3 \end{bmatrix}\transp$
2021-09-02 11:31:08 +09:00
which are the coordinates of the initial condition $\bm{u}(0)=\bm{u}_0$ in terms of
the eigenvectors $X$.
\begin{equation*}
\bm{u}_0 = c_1 \bm{x}_1 + c_2 \bm{x}_2 + c_3 \bm{x}_3
\end{equation*}
\begin{equation*}
\bm{c} =
\begin{bmatrix}
c_1\\
c_2\\
c_3
\end{bmatrix} = X^{-1} \bm{u}_0
\end{equation*}
and the general solution of the two equations are:
\begin{align*}
\bm{u}(t) &= e^{At} \bm{u}_0 = X e^{\Lambda t} X^{-1} \bm{u_0} &= X e^{\Lambda t} \bm{c} &= c_1 e^{\lambda_1 t} \bm{x}_1 + c_2 e^{\lambda_2 t} \bm{x}_2 + c_3 e^{\lambda_3 t} \bm{x}_3\\
\bm{u}_n &= A^n \bm{u}_0 = X \Lambda^n X^{-1} \bm{u_0} &= X \Lambda^n \bm{c} &= c_1 \lambda_1^n \bm{x}_1 + c_2 \lambda_2^n \bm{x}_2 + c_3 \lambda_3^n \bm{x}_3
2021-09-02 11:31:08 +09:00
\end{align*}
2022-11-06 20:51:41 +09:00
See Figure 9: Pattern 3 (P3) above again to get $XDc$.
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
\includegraphics[scale=0.8]{Pattern4.eps}
2021-09-02 11:31:08 +09:00
\caption{Pattern 4 - (P4)}
\end{figure}
This pattern (P4) works in both eigenvalue decomposition and singular value decomposition.
Both decompositions are expressed as a product of three matrices with a diagonal matrix in the middle,
and also a sum of rank 1 matrices with the eigenvalue/singular value coefficients.
More details are discussed in the next section.
\clearpage
\section{The Five Factorizations of a Matrix}
\begin{itemize}
\item Preface p.vii, The Plan for the Book.
\end{itemize}
2023-09-10 17:03:13 -04:00
$A=CR, A=LU, A=QR, A=Q \Lambda Q\transp, A=U \Sigma V\transp$ are
2021-09-02 11:31:08 +09:00
illustrated one by one.
2022-10-15 15:56:30 +09:00
\begin{table}[h]
\begin{tabular}{lll}
\Large{\boldmath $A=CR$} & \includegraphics{A_CR.eps} &
\begin{tabular}{l}
Independent columns in $C$\\
Row echelon form in $R$\\
Leads to column rank = row rank
\end{tabular}\\
\Large{\boldmath $A=LU$} & \includegraphics{A_LU.eps} &
\begin{tabular}{l}
$LU$ decomposition from\\
Gaussian elimination\\
(Lower triangular)(Upper triangular)
\end{tabular}\\
\Large{\boldmath $A=QR$} & \includegraphics{A_QR.eps} &
\begin{tabular}{l}
$QR$ decomposition as\\
Gram-Schmidt orthogonalization\\
Orthogonal $Q$ and triangular $R$
\end{tabular}\\
2023-09-10 17:03:13 -04:00
2022-10-15 15:56:30 +09:00
\Large{\boldmath $S=Q\Lambda Q\transp$} & \includegraphics{A_QLQT.eps} &
\begin{tabular}{l}
Eigenvalue decomposition\\
of a symmetric matrix $S$\\
Eigenvectors in $Q$, eigenvalues in $\Lambda$
\end{tabular}\\
2023-09-10 17:03:13 -04:00
2022-10-15 15:56:30 +09:00
\Large{\boldmath $A=U\Sigma V\transp$} & \includegraphics{A_USVT.eps} &
\begin{tabular}{l}
Singular value decomposition\\
of all matrices $A$\\
Singular values in $\Sigma$
\end{tabular}
\end{tabular}
\caption{The Five Factorization}
\end{table}
2021-09-02 11:31:08 +09:00
\subsection{$\boldsymbol{A=CR}$}
\begin{itemize}
\item Sec.1.4 Matrix Multiplication and $\bm{A=CR}$ (p.29)
\end{itemize}
2023-09-10 17:03:13 -04:00
The row rank and the column rank of a general rectangular matrix $A$ are equal.
2021-09-02 11:31:08 +09:00
This factorization is the most intuitive way to understand this theorem.
$C$ consists of independent columns of $A$, and $R$ is the row reduced echelon form of $A$.
$A=CR$ reduces to $r$ independent columns in $C$ times $r$ independent rows in $R$.
\begin{equation*}
\begin{split}
A &= CR\\
\begin{bmatrix}
1 & 2 & 3 \\
2 & 3 & 5
\end{bmatrix}
& =
\begin{bmatrix}
1 & 2 \\
2 & 3
\end{bmatrix}
\begin{bmatrix}
1 & 0 & 1 \\
0 & 1 & 1
\end{bmatrix}
\end{split}
\end{equation*}
Procedure: Look at the columns of $A$ from left to right. Keep independent ones,
discard dependent ones which can be created by the former columns.
The column 1 and the column 2 survive, and the column 3 is discarded
because it is expressed as a sum of the former two columns.
2023-09-10 17:03:13 -04:00
To rebuild $A$ by the independent columns 1 and 2, you find a row echelon form $R$
appearing on the right.
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
\includegraphics[scale=0.8]{CR1.eps}
2021-09-02 11:31:08 +09:00
\caption{Column Rank in $CR$}
\end{figure}
2023-09-10 17:03:13 -04:00
Now the column rank is two because there are only two independent columns in $C$
2021-09-02 11:31:08 +09:00
and all the columns of $A$ are linear combinations of the two columns of $C$.
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
\includegraphics[scale=0.8]{CR2.eps}
2021-09-02 11:31:08 +09:00
\caption{Row Rank in $CR$}
\end{figure}
2023-09-10 17:03:13 -04:00
And the row rank is also two because there are only two independent rows in $R$
2021-09-02 11:31:08 +09:00
and all the rows of $A$ are linear combinations of the two rows of $R$.
\subsection{$\boldsymbol{A=LU}$}
2023-09-10 17:03:13 -04:00
Solving $A\bm{x}=\bm{b}$ via Gaussian elimination can be represented as an $LU$ factorization.
Usually, you apply elementary row operation matrices ($E$) to $A$ to make upper triangular $U$.
2021-09-02 11:31:08 +09:00
\begin{align*}
EA &= U\\
A &= E^{-1}U\\
\text{let} \; L = E^{-1}, \quad A &= LU
\end{align*}
2022-10-15 14:45:45 +09:00
Now solve $A\bm{x}=\bm{b}$ in 2 steps: (1) forward $L\bm{c}=\bm{b}$ and (2) back $U\bm{x}=\bm{c}$.
2021-09-02 11:31:08 +09:00
\begin{itemize}
\item Sec.2.3 (p.57) Matrix Computations and $\bm{A=LU}$
\end{itemize}
Here, we directly calculate $L$ and $U$ from $A$.
\begin{equation*}
2023-09-10 17:03:13 -04:00
A =
2021-09-02 11:31:08 +09:00
\begin{bmatrix}
|\\
\bm{l}_1\\
|
\end{bmatrix}
\begin{bmatrix}
- \bm{u}^*_1 -
\end{bmatrix}
+ \begin{bmatrix}
0 & \begin{matrix} 0 & 0 \end{matrix}\\
\begin{matrix} 0 \\ 0 \end{matrix} & A_2
\end{bmatrix}
2023-09-10 17:03:13 -04:00
=
2021-09-02 11:31:08 +09:00
\begin{bmatrix}
|\\
\bm{l}_1\\
|
\end{bmatrix}
\begin{bmatrix}
- \bm{u}^*_1 -
\end{bmatrix}
+
\begin{bmatrix}
|\\
\bm{l}_2\\
|
\end{bmatrix}
\begin{bmatrix}
- \bm{u}^*_2 -
\end{bmatrix}
+ \begin{bmatrix}
0 & 0 & 0\\
0 & 0 & 0 \\
0 & 0 & A_3
\end{bmatrix} = LU
\end{equation*}
2023-09-10 17:03:13 -04:00
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
\includegraphics[scale=0.8]{LU1.eps}
2021-09-02 11:31:08 +09:00
\caption{Recursive Rank 1 Matrix Peeling from $A$}
\end{figure}
To find $L$ and $U$, peel off the rank 1 matrix made of
the first row and the first column of $A$.
This leaves $A_2$. Do this recursively and decompose $A$ into the sum of rank 1 matrices.
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
\includegraphics[scale=0.8]{LU2.eps}
2021-09-02 11:31:08 +09:00
\caption{$LU$ rebuilds $A$}
\end{figure}
To rebuild $A$ from $L$ times $U$, use column-row multiplication.
2021-09-02 11:31:08 +09:00
\subsection{$\boldsymbol{A=QR}$}
$A=QR$ changes the columns of $A$ into perpendicular columns of $Q$, keeping $\bm{C}(A) = \bm{C}(Q)$.
\begin{itemize}
\item Sec.4.4 Orthogonal matrices and Gram-Schmidt (p.165)
\end{itemize}
In Gram-Schmidt, the normalized $\bm{a}_1$ is $\bm{q}_1$.
Then $\bm{a}_2$ is adjusted to be perpendicular to $\bm{q}_1$ to create $\bm{q}_2$.
This procedure gives:
2021-09-02 11:31:08 +09:00
\begin{align*}
2022-10-01 08:30:37 +09:00
\bm{q}_1 &= \bm{a}_1/||\bm{a}_1|| \\
\bm{q}_2 &= \bm{a}_2 - (\bm{q}_1\transp \bm{a}_2)\bm{q}_1 , \quad \bm{q}_2 = \bm{q}_2/||\bm{q}_2|| \\
\bm{q}_3 &= \bm{a}_3 - (\bm{q}_1\transp \bm{a}_3)\bm{q}_1 - (\bm{q}_2\transp \bm{a}_3)\bm{q}_2, \quad \bm{q}_3 = \bm{q}_3/||\bm{q}_3||
2021-09-02 11:31:08 +09:00
\end{align*}
2023-09-10 17:03:13 -04:00
In the reverse direction, let $r_{ij} = \bm{q}_i\transp \bm{a}_j$ and you will get:
2021-09-02 11:31:08 +09:00
\begin{align*}
\bm{a}_1 &= r_{11}\bm{q}_1\\
\bm{a}_2 &= r_{12}\bm{q}_1 + r_{22} \bm{q}_2\\
\bm{a}_3 &= r_{13}\bm{q}_1 + r_{23} \bm{q}_2 + r_{33} \bm{q}_3
\end{align*}
The original $A$ becomes $QR$: orthogonal $Q$ times upper triangular $R$.
2021-09-02 11:31:08 +09:00
\begin{gather*}
2023-09-10 17:03:13 -04:00
A =
2021-09-02 11:31:08 +09:00
\begin{bmatrix}
| & | & |\\
\bm{q}_1 & \bm{q}_2 & \bm{q}_3\\
| & | & |
\end{bmatrix}
\begin{bmatrix}
r_{11} & r_{12} & r_{13}\\
& r_{22} & r_{23}\\
& & r_{33}
\end{bmatrix} = QR\\
\\
2022-08-14 15:51:43 +09:00
Q Q\transp=Q\transp Q = I
2021-09-02 11:31:08 +09:00
\end{gather*}
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
\includegraphics[scale=0.8]{QR.eps}
2021-09-02 11:31:08 +09:00
\caption{$A=QR$}
\end{figure}
2023-09-10 17:03:13 -04:00
Each column vector of $A$ can be rebuilt from $Q$ and $R$.
2021-09-02 11:31:08 +09:00
See Pattern 1 (P1) again for the graphic interpretation.
2022-08-14 15:51:43 +09:00
\subsection{$\boldsymbol{S=Q \Lambda Q\transp}$}
2021-09-02 11:31:08 +09:00
All symmetric matrices $S$ must have real eigenvalues and orthogonal eigenvectors.
2023-09-10 17:03:13 -04:00
The eigenvalues are the diagonal elements of $\Lambda$ and the eigenvectors are in $Q$.
2021-09-02 11:31:08 +09:00
\begin{itemize}
2021-10-03 20:34:28 +09:00
\item Sec.6.3 (p.227) Symmetric Positive Definite Matrices
2021-09-02 11:31:08 +09:00
\end{itemize}
\begin{align*}
2022-08-14 15:51:43 +09:00
S = Q \Lambda Q\transp
2021-09-02 11:31:08 +09:00
&= \begin{bmatrix}
| & | & |\\
\bm{q}_1 & \bm{q}_2 & \bm{q}_3\\
| & | & |
\end{bmatrix}
\begin{bmatrix}
\lambda_1 \\
& \lambda_2 & \\
& & \lambda_3
\end{bmatrix}
\begin{bmatrix}
2022-08-14 15:51:43 +09:00
- \bm{q}_1\transp -\\
- \bm{q}_2\transp -\\
- \bm{q}_3\transp -
2021-09-02 11:31:08 +09:00
\end{bmatrix}\\
\\
&=
\lambda_1 \begin{bmatrix}
|\\
\bm{q}_1\\
|
\end{bmatrix}
\begin{bmatrix}
2023-09-10 17:03:13 -04:00
- \bm{q}_1\transp -
2021-09-02 11:31:08 +09:00
\end{bmatrix}
+
\lambda_2 \begin{bmatrix}
|\\
\bm{q}_2\\
|
\end{bmatrix}
\begin{bmatrix}
2022-08-14 15:51:43 +09:00
- \bm{q}_2\transp -
2023-09-10 17:03:13 -04:00
\end{bmatrix}
2021-09-02 11:31:08 +09:00
+
\lambda_3 \begin{bmatrix}
|\\
\bm{q}_3 \\
|
\end{bmatrix}
\begin{bmatrix}
2022-08-14 15:51:43 +09:00
- \bm{q}_3\transp -
2021-09-02 11:31:08 +09:00
\end{bmatrix} \\
&= \lambda_1 P_1 + \lambda_2 P_2 + \lambda_3 P_3
\end{align*}
\begin{equation*}
2022-08-14 15:51:43 +09:00
P_1=\bm{q}_1 \bm{q}_1\transp, \quad P_2=\bm{q}_2 \bm{q}_2\transp, \quad P_3=\bm{q}_3 \bm{q}_3\transp
2021-09-02 11:31:08 +09:00
\end{equation*}
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
\includegraphics[scale=0.8]{EVD.eps}
2022-08-14 15:51:43 +09:00
\caption{$S=Q \Lambda Q\transp$}
2021-09-02 11:31:08 +09:00
\end{figure}
A symmetric matrix $S$ is diagonalized into $\Lambda$ by an orthogonal matrix $Q$
2022-08-14 15:51:43 +09:00
and its transpose. And it is broken down into a combination of rank 1 projection matrices $P=qq\transp$.
This is the spectral theorem.
2021-09-02 11:31:08 +09:00
Note that Pattern 4 (P4) is working for the decomposition.
\begin{gather*}
2022-08-14 15:51:43 +09:00
S=S\transp = \lambda_1 P_1 + \lambda_2 P_2 + \lambda_3 P_3\\
QQ\transp = P_1 + P_2 + P_3 = I \\
2021-09-02 11:31:08 +09:00
P_1 P_2 = P_2 P_3 = P_3 P_1 = O\\
2022-08-14 15:51:43 +09:00
P_1^2 =P_1=P_1\transp, \quad P_2^2=P_2=P_2\transp, \quad P_3^2=P_3=P_3\transp
2021-09-02 11:31:08 +09:00
\end{gather*}
2022-08-14 15:51:43 +09:00
\subsection{$\boldsymbol{A=U \Sigma V\transp}$}
2021-09-02 11:31:08 +09:00
\begin{itemize}
2023-07-18 10:18:41 +08:00
\item Sec.7.1 (p.259) Singular Values and Singular Vectors
2021-09-02 11:31:08 +09:00
\end{itemize}
Every matrix (including rectangular one) has a singular value decomposition (SVD).
2022-10-15 14:45:45 +09:00
$A=U \Sigma V\transp$ has the singular vectors of $A$ in $U$ and $V$.
2023-09-10 17:03:13 -04:00
The following figure illustrates the 'reduced' SVD.
2022-10-15 14:45:45 +09:00
2021-09-02 11:31:08 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
\includegraphics[scale=0.8]{SVD.eps}
2022-08-14 15:51:43 +09:00
\caption{$A=U \Sigma V\transp$}
2021-09-02 11:31:08 +09:00
\end{figure}
2023-09-10 17:03:13 -04:00
You can find $V$ as an orthonormal basis of $\mathbb{R}^n$ (eigenvectors of $A\transp A$)
2022-08-14 15:51:43 +09:00
and $U$ as an orthonormal basis of $\mathbb{R}^m$ (eigenvectors of $AA\transp$).
2021-09-02 11:31:08 +09:00
Together they diagonalize $A$ into $\Sigma$.
2023-09-10 17:03:13 -04:00
This can be also expressed as a combination of rank 1 matrices.
2021-09-02 11:31:08 +09:00
\begin{align*}
2022-08-14 15:51:43 +09:00
A = U \Sigma V\transp =
2021-09-02 11:31:08 +09:00
\begin{bmatrix}
| & | & |\\
\bm{u}_1 & \bm{u}_2 & \bm{u}_3\\
| & | & |
\end{bmatrix}
\begin{bmatrix}
\sigma_1 \\
& \sigma_2 \\
& &
\end{bmatrix}
\begin{bmatrix}
2022-08-14 15:51:43 +09:00
- \bm{v}_1\transp -\\
- \bm{v}_2\transp -
2021-09-02 11:31:08 +09:00
\end{bmatrix}
& =
\sigma_1 \begin{bmatrix}
|\\
\bm{u}_1\\
|
\end{bmatrix}
\begin{bmatrix}
2023-09-10 17:03:13 -04:00
- \bm{v}_1\transp -
2021-09-02 11:31:08 +09:00
\end{bmatrix}
+
\sigma_2 \begin{bmatrix}
|\\
\bm{u}_2\\
|
\end{bmatrix}
\begin{bmatrix}
2022-08-14 15:51:43 +09:00
- \bm{v}_2\transp -
2021-09-02 11:31:08 +09:00
\end{bmatrix} \\
2022-08-14 15:51:43 +09:00
& = \sigma_1 \bm{u}_1 \bm{v}_1\transp + \sigma_2 \bm{u}_2 \bm{v}_2\transp
2021-09-02 11:31:08 +09:00
\end{align*}
Note that:
\begin{align*}
2022-08-14 15:51:43 +09:00
U U\transp &= I_m \\
V V\transp &= I_n
2021-09-02 11:31:08 +09:00
\end{align*}
See Pattern 4 (P4) for the graphic notation.
\section*{Conclusion and Acknowledgements}
2023-09-10 17:03:13 -04:00
I have presented a systematic visualization of matrix/vector multiplication and
its applications to the Five Matrix Factorizations. I hope you
enjoy them and find them useful
in understanding Linear Algebra.
2021-09-02 11:31:08 +09:00
2023-09-10 17:03:13 -04:00
Ashley Fernandes helped me with typesetting, which
makes this paper much more appealing and professional.
2021-09-02 11:31:08 +09:00
To conclude this paper, I'd like to thank Prof. Gilbert Strang for
2023-09-10 17:03:13 -04:00
publishing ``Linear Algebra for Everyone". It presents a new pathway to these beautiful landscapes in Linear Algebra.
2021-09-02 11:31:08 +09:00
Everyone can reach a fundamental understanding of its underlying ideas
in a practical manner that introduces us to contemporary and also
2023-09-12 16:58:18 +09:00
traditional data science and machine learning.
2021-09-02 11:31:08 +09:00
2021-10-01 16:44:25 +09:00
\section*{References and Related Works}
\begin{enumerate}
2023-09-10 17:03:13 -04:00
\item
2021-10-03 20:34:28 +09:00
Gilbert Strang(2020),\emph{Linear Algebra for Everyone}, Wellesley Cambridge Press.,\\
2023-09-12 16:58:18 +09:00
\url{http://math.mit.edu/everyone}
2021-10-01 16:44:25 +09:00
\item
2023-09-12 16:58:18 +09:00
Gilbert Strang(2016), \emph{Introduction to Linear Algebra},Wellesley Cambridge Press, 6th ed.,\\
\url{http://math.mit.edu/linearalgebra}
2023-03-23 18:45:15 +09:00
\item Kenji Hiranabe(2021), \emph{Map of Eigenvalues}, Slidedeck,\\
2023-09-12 16:58:18 +09:00
\url{https://github.com/kenjihiranabe/The-Art-of-Linear-Algebra/blob/main/MapofEigenvalues.pdf}
2022-08-13 12:37:44 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
2022-08-13 12:37:44 +09:00
\includegraphics[keepaspectratio, width=\linewidth]{MapofEigenvalues.eps}
\caption{Map of Eigenvalues}
\end{figure}
2023-03-23 18:45:15 +09:00
\item Kenji Hiranabe(2020), \emph{Matrix World}, Slidedeck,\\
\url{https://github.com/kenjihiranabe/The-Art-of-Linear-Algebra/blob/main/MatrixWorld.pdf}\\
2022-08-13 12:37:44 +09:00
\begin{figure}[H]
2022-10-15 15:56:30 +09:00
\centering
2022-08-13 12:37:44 +09:00
\includegraphics[keepaspectratio, width=\linewidth]{MatrixWorld.eps}
\caption{Matrix World}
\end{figure}
2023-03-23 18:45:15 +09:00
\item Gilbert Strang, artwork by Kenji Hiranabe, \emph{The Four Subspaces and the solutions to $A\bm{x}=\bm{b}$}\\
\begin{figure}[H]
2023-03-23 18:45:15 +09:00
\centering
2023-09-12 16:58:18 +09:00
\includegraphics[scale=0.6]{TheFourSubspaces.eps}
\caption{The Four Subspaces and the solutions to $A\bm{x}=\bm{b}$}
\end{figure}
2021-10-01 16:44:25 +09:00
\end{enumerate}
2023-08-22 09:07:48 +09:00
\end{document}